Abstract: Methods and systems directed to a user equipment accessing a wireless-communication cell are described. The methods and systems include a user equipment (110) selecting (805) a first wireless-communication cell (306) and receiving (815), from a base station (124) providing the first wireless-communication cell (306), a first system information block (SIB). Based on contents of an entry within the SIB that indicate a core network type of the first wireless-communication cell (306), the user equipment (110) selects (825) a second wireless-communication cell (304) and receives (835), from a second base station (122) providing the second wireless-communication cell (304), a second SIB.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for receiving, at a mobile computing device that is associated with a called user, a call from a calling computing device that is associated with a calling user; in response to receiving the call, determining, by the mobile computing device, that data associated with the called user indicates that the called user will not respond to the call; in response to determining that the called user will not respond to the call, inferring, by the mobile computing device, an informational need of the calling user; and automatically providing, from the mobile computing device to the calling computing device, information associated with the called user and that satisfies the inferred informational need of the calling user.

Abstract: The subject matter of this specification can be embodied in, among other things, a computer-implemented method that includes obtaining, at a computer system, a plurality of contact identifiers for a computer account holder, wherein the identifiers represent a plurality of different communication modes; identifying a handle for the account holder, wherein the handle is associated with a uniform resource locator; and correlating the handle with the plurality of contact identifiers, and storing the handle and plurality of contact identifiers together so as to permit retrieval of the contact identifiers in response to identification of the handle.

Abstract: The subject matter of this specification can be embodied in, among other things, a method that includes receiving an audio signal and initiating speech recognition tasks by a plurality of speech recognition systems (SRS's). Each SRS is configured to generate a recognition result specifying possible speech included in the audio signal and a confidence value indicating a confidence in a correctness of the speech result. The method also includes completing a portion of the speech recognition tasks including generating one or more recognition results and one or more confidence values for the one or more recognition results, determining whether the one or more confidence values meets a confidence threshold, aborting a remaining portion of the speech recognition tasks for SRS's that have not generated a recognition result, and outputting a final recognition result based on at least one of the generated one or more speech results.

Abstract: A method for data storage includes assigning multiple different portions of a given object, which is stored in non-volatile storage, for access by multiple respective software modules running on one or more processors. Storage operations are performed by the software modules in the portions of the given object. Respective local values, which are indicative of most recent times the storage operations were performed in the given object by the respective software modules, are updated by the software modules. A global value is updated for the given object, by consolidating updates of the local values performed by the software modules.

Abstract: A method includes obtaining a plaintext query that includes a sequence of plaintext integers and generating a polynomial having coefficients that include the sequence of plaintext integers of the plaintext query. The method also includes encrypting the polynomial using a secret encryption key and transmitting the encrypted polynomial to a server. The secret encryption key is randomly sampled from a ciphertext space and the server is configured to expand the encrypted polynomial using a public encryption key to obtain a sequence of encrypted integers corresponding to the sequence of plaintext integers. The method also includes receiving an encrypted result from the server. The encrypted result is based on the sequence of encrypted integers.

Abstract: Methods, systems, and media for identifying video objects linked to a source video are provided. In some embodiments, the method comprises: identifying demographic attributes corresponding to a first user participating in an online conversation; determining at least one keyword associated with the online conversation, wherein the keyword indicates a topic of the online conversation; identifying a video object based at least on the demographic attributes and the at least one keyword, wherein the video object comprises a portion of a video; causing the identified video object to be presented in a group of video objects on a first user device associated with the first user; receiving an indication that the identified video object has been selected on the first user device for inclusion in a message in the online conversation; and causing the identified video object to be presented on a second user device associated with the second user.

Abstract: Techniques and apparatuses are described that implement a smart-device-based radar system capable of performing angular estimation using machine learning. In particular, a radar system 102 includes an angle-estimation module 504 that employs machine learning to estimate an angular position of one or more objects (e.g., users). By analyzing an irregular shape of the radar system 102's spatial response across a wide field of view, the angle-estimation module 504 can resolve angular ambiguities that may be present based on the angle to the object or based on a design of the radar system 102 to correctly identify the angular position of the object. Using machine-learning techniques, the radar system 102 can achieve a high probability of detection and a low false-alarm rate for a variety of different antenna element spacings and frequencies.

Abstract: A computing system receives requests from client devices to process voice queries that have been detected in local environments of the client devices. The system identifies that a value that is based on a number of requests to process voice queries received by the system during a specified time interval satisfies one or more criteria. In response, the system triggers analysis of at least some of the requests received during the specified time interval to trigger analysis of at least some received requests to determine a set of requests that each identify a common voice query. The system can generate an electronic fingerprint that indicates a distinctive model of the common voice query. The fingerprint can then be used to detect an illegitimate voice query identified in a request from a client device at a later time.

Abstract: A method includes receiving audio data encoding an utterance and obtaining a set of bias phrases corresponding to a context of the utterance. Each bias phrase includes one or more words. The method also includes processing, using a speech recognition model, acoustic features derived from the audio to generate an output from the speech recognition model. The speech recognition model includes a first encoder configured to receive the acoustic features, a first attention module, a bias encoder configured to receive data indicating the obtained set of bias phrases, a bias encoder, and a decoder configured to determine likelihoods of sequences of speech elements based on output of the first attention module and output of the bias attention module. The method also includes determining a transcript for the utterance based on the likelihoods of sequences of speech elements.

Abstract: A method includes receiving audio data encoding an utterance, processing, using a speech recognition model, the audio data to generate speech recognition scores for speech elements, and determining context scores for the speech elements based on context data indicating a context for the utterance. The method also includes executing, using the speech recognition scores and the context scores, a beam search decoding process to determine one or more candidate transcriptions for the utterance. The method also includes selecting a transcription for the utterance from the one or more candidate transcriptions.

Abstract: Systems and methods of generating voice-based software applications are provided. A system can receive, from an application developer computing device, a request to build a voice-based software application. The system can select an application template from a plurality of application templates. The selected application template can include a module that corresponds to a function of the voice-based software application. The system can provide the selected application template to the application developer computing device. The system can receive, from the application developer computing device, an input for a field of the at least one module of the selected application template. The system can generate the voice-based software application based on the selected application template and the input for the at least one field of the at least one module of the selected application template.

Abstract: A user-friendly programmable thermostat is described that includes a body having a central electronic display surrounded by a ring that can be rotated and pressed inwardly to provide user input in a simple and elegant fashion. The current temperature and setpoint temperature are graphically displayed as prominent tick marks over a range of background tick marks on the electronic display. Different colors can be displayed to indicate currently active HVAC functions, and different intensities of colors can be displayed to indicate an amount of heating or cooling required to reach a target temperature. The setpoint temperature for the device can be altered by user rotation of the rotatable ring, and the programmed schedule can be displayed to the user and altered by the user by virtue of rotations and inward pressings of the ring.

Abstract: Techniques and apparatuses are described that enable radar attenuation mitigation. To improve radar performance, characteristics of an attenuator and/or properties of a radar signal are determined to reduce attenuation of the radar signal due to the attenuator and enable a radar system to detect a target located on an opposite side of the attenuator. These techniques are beneficial in situations in which the attenuator is unavoidably located between the radar system and a target, either due to integration within other electronic devices or due to an operating environment. These techniques save power and cost by reducing the attenuation without increasing transmit power or changing material properties of the attenuator.

Abstract: A method of receiving EM field magnitude values indicative of a first pose of a mobile unit in relation to a base unit, receiving sensor data from a second sensor associated with the mobile unit, where the sensor data is indicative of a direction of movement of the mobile unit, calculating a set of candidate pose solutions based on the EM field magnitude values, selecting a pose from the set of candidate pose solutions based on the sensor data from the second sensor, and sending the pose to the processor.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for cross input modality learning in a mobile device are disclosed. In one aspect, a method includes activating a first modality user input mode in which user inputs by way of a first modality are recognized using a first modality recognizer; and receiving a user input by way of the first modality. The method includes, obtaining, as a result of the first modality recognizer recognizing the user input, a transcription that includes a particular term; and generating an input context data structure that references at least the particular term. The method further includes, transmitting, by the first modality recognizer, the input context data structure to a second modality recognizer for use in updating a second modality recognition model associated with the second modality recognizer.

Abstract: Wearable electronic devices having a selfie camera are described herein. The wearable electronic devices include a body defining an aperture that is sized and shaped to receive a finger of a user. The wearable electronic device also includes a processor housed in the body, an input device arranged on an outer surface of the body and communicatively coupled to the processor to receive an input from a second finger of the user, an optical sensor to capture an image in response to the input from the second finger of the user, a transmitter communicatively coupled to the processor to transmit the image to an external electronic device, and optionally a power source communicatively coupled to the processor and the transmitter to provide power to at least the processor and the transmitter.

Abstract: Methods and apparatus related to determining a ranking of class members of a task component class. A task component class identifier may be identified, the task component class identifier having a plurality of associated class members. One or more task completion indicators having the identified at least one task component class identifier may be determined. Each of the task completion indicators may indicate which of the associated class members was utilized to complete a task. One or more of the associated class members for the task component class identifier may be ranked. The ranking may be based on frequency of occurrence of the given associated class member in the task completion indicators.

Abstract: A quantum computing device includes: a first array of qubits arranged along a first axis; and a second array of qubits arranged along a second axis different from the first axis so that the qubits of the second array intersect with the qubits of the first array to form a lattice structure, in which each qubit in the first array is offset along the second axis relative to a directly adjacent qubit in the first array, each qubit in the second array is offset along the first axis relative to a directly adjacent qubit in the second array, and each intersection between a qubit from the first array and a qubit from the second array in the lattice structure comprises a coupler arranged to inductively couple the qubit from the first array to the qubit from the second array.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for an object feature identification system employed by a robotic are disclosed. In one aspect, a method includes the actions of generating a data reading of a work area by scanning the work area with a sensor device of the robot; identifying, by processing the data reading through a learning engine, a particular component of a plurality of components associated with the work area based on a task to be performed; identifying, with the machine learning engine, a particular feature of the particular component used in a completion of the task; determining, with the machine learning engine, a particular tool of a plurality of tools of the robot that is configured to perform the task; and performing the task with the particular tool and the particular feature of the particular component.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for using location aliases. In some implementations, a query is received, and a user that submitted the query is identified. It can be determined that one or more terms of the query represent an alias for a user-specific geographical location that has not been designated for the identified user. In response, a prompt can be provided to the user to specify a geographical location corresponding to the one or more query terms, data indicating a geographical location is received, and data is stored that indicates that, for the identified user, the one or more terms are an alias for the geographical location. One or more search results responsive to the query are provided, where the alias corresponds to the geographical location input in response to the prompt.

Abstract: Replicas are selected in a large distributed network, and the roles for these replicas are identified. In one example, a leader is selected from among candidate computing clusters. To make this selection, an activity monitor predicts or monitors the workload of one or more clients. Different activities of the workload are given corresponding weights. The delay in performing requested activities, modified by these weights is found, and the candidate leader with the lowest weighted delay is selected as the leader.

Abstract: Example aspects of the present disclosure are directed to systems and methods that employ a machine-learned opinion classification model to classify portions (e.g., sentences, phrases, paragraphs, etc.) of documents (e.g., news articles, web pages, etc.) as being opinions or not opinions. Further, in some implementations, portions classified as opinions can be considered for inclusion in an informational display. For example, document portions can be ranked according to importance and selected for inclusion in an informational display based on their ranking. Additionally or alternatively, for systems which access and consider multiple documents, the portions of a document that are classified as opinion can be compared to similarly-classified portions of other documents to perform document clustering, to ensure diversity within a presentation, and/or other tasks.

Abstract: In one example, a computing device includes one or more user input detection components, and one or more processors configured to receive an indication of a first user input detected by the one or more user input detection components, responsive to receiving the indication of the first user input, adjust a level of an attention buffer at a defined rate; responsive to determining that the level of the attention buffer satisfies a first threshold, prevent further interaction with a user interface of the computing device, responsive to determining that an indication of a second user input has not been received within a time period, adjust a level of the attention buffer, and responsive to determining that the level of the attention buffer satisfies a second threshold, allow further interaction with the user interface.

Abstract: This document describes methods, devices, systems, and means for an active coordination set for mobility management. A user equipment (110) evaluates a link quality measurement for one or more base stations (120) and determines to include at least a first base station of the base stations (120) in an Active Coordination Set (ACS). The user equipment (110) sends a message, including an indication to add the at least first base station (120) to the ACS, to an ACS Server (520) that causes the ACS Server (520) to store the ACS for the user equipment (110) and send a copy of the stored ACS to a master base station (121). The user equipment (110) communicates via one or more of the base stations (120) included in the ACS. The user equipment (110) can transmit an uplink ACS sounding signal to evaluate base stations (120) to include in the ACS.

Abstract: Methods, systems, and apparatus, including computer programs encoded on computer storage media, for generating descriptions of input images. One of the methods includes obtaining an input image; processing the input image using a first neural network to generate an alternative representation for the input image; and processing the alternative representation for the input image using a second neural network to generate a sequence of a plurality of words in a target natural language that describes the input image.

Abstract: Mechanisms for managing output of an HDMI source are provided. In accordance with some implementations of the disclosed subject matter, a method for controlling output of an HDMI source is provided, the method comprising: establishing a connection between the HDMI source and an HDMI sink at a first address of a consumer electronic control bus of the HDMI sink; sending a request for an identity of the active source connected to the HDMI sink; monitoring signals on the consumer electronic control bus; receiving a message over the consumer electronic control bus identifying a second address on the consumer electronic control bus different from the first address as an address of an active source; setting a status of the HDMI source as inactive in response to receiving the message; and inhibiting output of video from the HDMI source to the HDMI sink in response to the status being set as inactive.

Abstract: Methods, systems, and apparatus for performing a matrix multiplication using a hardware circuit are described. An example method begins by obtaining an input activation value and a weight input value in a first floating point format. The input activation value and the weight input value are multiplied to generate a product value in a second floating point format that has higher precision than the first floating point format. A partial sum value is obtained in a third floating point format that has a higher precision than the first floating point format. The partial sum value and the product value are combined to generate an updated partial sum value that has the third floating point format.

Abstract: Techniques (300, 400, 500) and apparatuses (100, 200, 700) for recognizing accented speech are described. In some embodiments, an accent module recognizes accented speech using an accent library based on device data, uses different speech recognition correction levels based on an application field into which recognized words are set to be provided, or updates an accent library based on corrections made to incorrectly recognized speech.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for receiving, at a search engine, a search query submitted by a user to the search engine; identifying an entity that is associated with the search query; determining that the user that submitted the search query to the search engine is associated with the entity in an entity listing portal; and in response to determining that the user that submitted the search query to the search engine is associated with the entity in an entity listing portal, providing a search results webpage that includes i) search results that the search engine generated responsive to the search query and ii) an interface through which edits to data associated with the entity within the entity listing portal can be provided.

Abstract: Systems and methods according to aspects of the invention verify requests for information over a network. Computing devices communicating over a network transmit numerous signals with various types of information including spatial information, temporal information, device-specific information, and transmission specific information. This information may be used to generate statistical records of valid request for information. Various combinations of these signals may be used to determine whether a request for information is a valid request based on the statistical records associated with the signals. Where a request for information has been verified, the system and method may provide the information. Where the request for information has not been verified, the system and method may deny the request for information. Thus the system and method may be used to prevent and block scraping and spamming attacks.

Abstract: Systems and techniques are provided for determining mutual information with absolute dependency for feature selection. Items may be received from a dataset. Each item may include two random variables. A first random variable may be associated with a first range of discrete values, and a second random variable may be associated with a second range of discrete values.

Abstract: Systems and methods are provided for sharing or recreating a state from a mobile device. For example, a mobile application state identifier can include a protocol identifier, a target task component associated with a mobile application executing when the application state identifier is generated and a sub-task component related to a user-interface active in the mobile application when the application state identifier is generated. The application state identifier can also include a user input component reflecting actions taken by a user during execution of the mobile application. A method can include generating, on a first mobile device, an application state identifier and transmitting the application state identifier to a second mobile device, wherein the second mobile device opens a mobile application associated with the target task to a user-interface corresponding to the sub-task component in response to receiving the application state identifier.

Abstract: A method for automatically enrolling a smart-home device in a demand-response program includes receiving an identification of a user account that is sent from a utility provider computer system based on an agreement that the smart-home device will be enrolled in the demand-response program. The identification of the user account is also sent to indicate to the device management server that the smart-home device should be sent to a location associated with the user account. The method also includes causing the smart-home device to be sent to the location associated with the user account. The method additionally includes receiving an indication from the smart-home device that the smart-home device has been installed at the location associated with the user account, and enrolling the smart-home device in the demand-response program.

Abstract: A method for coding a current block using an intra-prediction mode includes defining a mapping from available intra-prediction modes to intra-prediction classes; determining, using the mapping, a first intra-prediction class of a first intra-prediction mode used for decoding a first neighboring block of the current block; determining, using the mapping, a second intra-prediction class of a second intra-prediction mode used for decoding a second neighboring block of the current block; using the first intra-prediction class and the second intra-prediction class as indices into a list of available context models to select a context model for coding the intra-prediction mode; and coding the intra-prediction mode using the context model. A first number of the intra-prediction classes is smaller than a second number of the available intra-prediction modes. each class is an ordinal value, and each available intra-prediction mode uniquely maps to one class of the intra-prediction classes.

Abstract: A user device can send, to a server, a request for a set of documents likely to be opened by a user, determine a client-suggested document to present to the user and a potential motive for the user to open the client-suggested document, receive a suggestion message from the server, the suggestion message including a set of documents likely to be opened by the user and potential motives for the user to open documents in the set of documents, and present, on a display of the user device, visual representations of the client-suggested document, the potential motive for the user to open the client-suggested document, multiple documents included in the set of documents, and the potential motives for the user to open the multiple documents in the set of documents.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for language models using domain-specific model components. In some implementations, context data for an utterance is obtained. A domain-specific model component is selected from among multiple domain-specific model components of a language model based on the non-linguistic context of the utterance. A score for a candidate transcription for the utterance is generated using the selected domain-specific model component and a baseline model component of the language model that is domain-independent. A transcription for the utterance is determined using the score the transcription is provided as output of an automated speech recognition system.

Abstract: A fast cryptographic hash of an input file using multiplication and permutation operations in a parallel processing environment. An example method includes updating an internal state for each of a plurality of packets, the packets being read from an input file. Updating the state for a packet can include injecting the packet into an internal state, mixing the bits of the internal state using multiplication, and shuffling the result of the multiplication so that bits with highest quality are permuted to locations that will propagate most widely in a next multiplication operation. The method also includes performing a reduction on the internal state and repeating the update of the internal state, the reduction, and the injecting a second time. The method may further include finalizing the internal state and storing a portion of the final internal state as a cryptographic hash of the input file.

Abstract: An information system generates data, generates an audio token, associates the audio token with the data, and transmits the audio token along with the data to the account management system. The information system configures a broadcasting computing device to broadcast the audio token at a location. User computing devices at the location via an audio communication channel receive the broadcast token and retransmit the audio token to the account management system via a network. The account management system receives the audio token from a particular user computing device along with a request for data associated with the audio token, extracts the data associated with the audio token data, and transmits the data associated with the audio token to the particular user computing device.

Abstract: A device is described that, while operating in an automobile configuration, causes a vehicle system of an automobile to display part of an automobile graphical user interface (GUI) while locally displaying a different part of the automobile GUI. Responsive to receiving user input for switching to a non-automobile GUI that permits access to at least some features of the device that are restricted by the automobile GUI, the device replaces the part of the automobile GUI that is locally displayed on the device with the non-automobile GUI, all while continuing to cause the vehicle system to display its part of the automobile GUI. The device accesses the at least some features of the device and responsive to receiving user input for switching back to the automobile GUI, the device replaces the non-automobile GUI with the part of the automobile GUI that was previously displayed by the device.

Abstract: An actuator, attached to a panel extending in a plane, includes a frame and a magnetic circuit assembly that includes a magnet and a voice coil, the magnet and voice coil being moveable relative to each other during operation of the actuator along an axis perpendicular to the panel's plane. The actuator further includes one or more suspension members attaching the frame to a first component of the magnetic circuit assembly. Each suspension member includes a segment attaching the suspension member to the frame at an attachment point and a first arm extending away from the attachment point to an end attached to the first component of the magnetic circuit assembly. During operation of the actuator the first arm of the one or more suspension members flex to accommodate axial displacements of the magnet relative to the voice coil.

Abstract: Techniques and apparatuses are described that enable radar modulations for radar sensing using a wireless communication chipset. A controller initializes or controls modulations performed by the wireless communication chipset. In this way, the controller can enable the wireless communication chipset to perform modulations for wireless communication or radar sensing. In some cases, the controller can further select a wireless communication channel for setting a frequency and a bandwidth of a radar signal, thereby avoiding interference between multiple radar signals or between the radar signal and a communication signal. In other cases, the controller can cause the wireless communication chipset to modulate a signal containing communication data using a radar modulation. This enables another device that receives the signal to perform wireless communication or radar sensing. By utilizing these techniques, the wireless communication chipset can be used for wireless communication or radar sensing.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for assisting participation in a social network. In one aspect, a method is performed by a system of one or more data processing devices. The method includes receiving, at the system, a historical record of message exchange between an individual and members in a member network, the system determining, for each of the members, whether the individual is likely to want to be related to the respective member, each determination considering the number and transactional characteristics of the message exchange between the individual and the respective member in the historical record, and the system outputting the determinations that the individual is likely to want to be related to at least two of the respective members.

Abstract: This document describes methods, devices, systems, and means for a low-power connected mode for wireless communication systems. In aspects, a user equipment (110) detects a thermal state or a battery state-of-charge while operating in a Radio Resource Control (RRC) connected mode (302). Based on the thermal state or the battery state-of-charge, the user equipment (110) transmits a request message to a base station (120) to request to enter into a low-power connected mode (304). The user equipment (110) receives a configuration message from the base station (120) to activate the low-power connected mode (306). The user equipment (110) then activates the low-power connected mode (308). Activating the low-power connected mode enables the user equipment (110) to reduce the thermal state and power consumption by maintaining an RRC connection with reduced functionalities, such as without monitoring a downlink channel for downlink signals from the base station (120).

Abstract: Techniques and apparatuses are described that enable power-conserving cache memory usage. Main memory constructed using, e.g., DRAM can be placed in a low-power mode, such as a self-refresh mode, for longer time periods using the described techniques and apparatuses. A hierarchical memory system includes a supplemental cache memory operatively coupled between a higher-level cache memory and the main memory. The main memory can be placed in the self-refresh mode responsive to the supplemental cache memory being selectively activated. The supplemental cache memory can be implemented with a highly- or fully-associative cache memory that is smaller than the higher-level cache memory. Thus, the supplemental cache memory can handle those cache misses by the higher-level cache memory that arise because too many memory blocks are mapped to a single cache line. In this manner, a DRAM implementation of the main memory can be kept in the self-refresh mode for longer time periods.

Abstract: A method for generating final transcriptions representing numerical sequences of utterances in a written domain includes receiving audio data for an utterance containing a numeric sequence, and decoding, using a sequence-to-sequence speech recognition model, the audio data for the utterance to generate, as output from the sequence-to-sequence speech recognition model, an intermediate transcription of the utterance. The method also includes processing, using a neural corrector/denormer, the intermediate transcription to generate a final transcription that represents the numeric sequence of the utterance in a written domain. The neural corrector/denormer is trained on a set of training samples, where each training sample includes a speech recognition hypothesis for a training utterance and a ground-truth transcription of the training utterance. The ground-truth transcription of the training utterance is in the written domain.

Abstract: The present disclosure describes techniques and systems directed to determining an intended input to a user interface from detected gesture positions. The described techniques and systems include detecting positions associated with a gesture made relative the user interface, associating a timing profile to the detected positions, and determining, from the detected positions and the associated timing profile, the intended input to the user interface.

Abstract: The present disclosure provides a text display system that switches seamlessly between an overview mode and a detailed view mode, depending on passive signals from the user. Accordingly, users can switch between the overview and detailed context using interactions that they use naturally. When looking for interesting things to read, they can naturally scroll fast and be able to see only the important topics of the contents. When they slow down the scrolling, the content will be presented in its original form without highlighting, so the user will be able to consume the content in the true form that the author created. Since there is no active interaction required other than the natural scrolling that users would do anyway, there is no mental hurdle in switching between the overview and detailed contexts, allowing users to find and consume content quickly.

Abstract: A method includes receiving a spoken utterance that includes a plurality of words, and generating, using a neural network-based utterance classifier comprising a stack of multiple Long-Short Term Memory (LSTM) layers, a respective textual representation for each word of the of the plurality of words of the spoken utterance. The neural network-based utterance classifier trained on negative training examples of spoken utterances not directed toward an automated assistant server. The method further including determining, using the respective textual representation generated for each word of the plurality of words of the spoken utterance, that the spoken utterance is one of directed toward the automated assistant server or not directed toward the automated assistant server, and when the spoken utterance is directed toward the automated assistant server, generating instructions that cause the automated assistant server to generate a response to the spoken utterance.

Abstract: Provided are methods and systems for enhancing speech when corrupted by transient noise (e.g., keyboard typing noise). The methods and systems utilize a reference microphone input signal for the transient noise in a signal restoration process used for the voice part of the signal. A robust Bayesian statistical model is used to regress the voice microphone on the reference microphone, which allows for direct inference about the desired voice signal while marginalizing the unwanted power spectral values of the voice and transient noise. Also provided is a straightforward and efficient Expectation-maximization (EM) procedure far fast enhancement of the corrupted signal. The methods and systems are designed to operate easily in real-time on standard hardware, and have very low latency so that there is no irritating delay in speaker response.